Core dev meeting #80

Personal update (howo)

Posted MRs for API limitation (memory issue, low priority since drone-protected) and a fix for the recurring-transfer bug.

Rebasing Hivemind MRs.

Side project: a second HiveD implementation in go — replays up to 30M blocks before failing. Curious whether a fresh AI pass with full repo context could outperform the original. Plans to maybe run a top-100 witness on it for the experiment.

Building a HAF app to better support HBD recurring transfers and payment — exposes an endpoint to query whether a given invoice ID is being paid, payment failure counts, etc. (closes a UX gap where devs currently have to walk virtual ops).

Drafting a design doc for encrypted on-chain content (Patreon/Substack-style paid content paid in HBD).

Built a per community notification filter on hivemind

Hive core / release (blocktrades)

Tagging the release today (RC14 is the final candidate). Pre-replayed snapshot will be uploaded; replay times are much better now.

Stack split across two servers; monitoring setup is the last piece and is mostly done.

API node upgrades to coordinate, but stagger them — don't have everyone switch at once unless they have backups.

Big changes in this release:

Major haf schema rewrite + many haf apps rewritten to match.
Hivemind near-complete rewrite — almost entirely SQL now, very little Python left.
Account-balance refactor (maintenance), build-time optimizations, include-file cleanup — held back to adjacent branches until after release.
Found new deadlock + index-creation bugs in haf.
Specialized exceptions added to wax (parseable error handling for apps).
Denser progressing well; will publicize as alternate to hive.blog and beef up its server.
Python WorkerBee underway, mirroring the TypeScript version.

Haf apps / Block Explorer (mcfarhat)

New endpoints on haf block explorer: historical proposal votes per proposal, total wallet addresses (daily granularity), proxy-account sorting, proxy-power API fixes, balance tracker fix for negative savings withdrawals.

Two new projects spun up: haf-stats and haf-fyp.

FYP: TikTok-style personalized ranking using Hivesense, votes/comments/community subs as weighted signals; open source, hosted on haf API nodes.
haf-stats: financial summaries, RC footprint per account, network-wide RC utilization, HP distribution buckets, and many more endpoints.

Repos created on GitLab but he can't push — needs maintainer permission from blocktrades/Gandalf. Action: blocktrades/Gandalf to grant.

Block-explorer frontend: TVL card, incoming/outgoing RC delegations with timestamps, new per-account balance breakdown (open orders, pending conversions, savings under withdrawal, escrow), Keychain + Hive Signer login, customizable dashboard, tooltip/crash fixes.

Pending MRs (4): historical proposal votes, total wallet addresses, proxy-account sorting, proxy-API fixes — Michael to review this week.

howo to share his Hivemind ranking-algorithm MR with mcfarhat for cross-pollination on the FYP algorithm.

Velocity L2

Very productive month: ~1.5M lines, ~2000 commits across 12 repos.

Built an inbox/outbox coordination system between 12 parallel Claude (Opus 4.7) sessions after team-agents broke down past 3–4 agents. Hits weekly Opus limit at ~14h of session time.

Velocity = new L2 with three pillars:

Block speed: 0.5 ms via RocksDB + heavy compression — Hive's 3s and Magi's 30s aren't viable for fast-paced game reward loops.
SDKs for Unity, Unreal, Phaser.js, GDScript — covers PS5/Xbox/Switch/mobile/PC so indie devs don't need to learn 4–5 new languages to ship.
Namespace isolation — each game gets its own fuel allocation and isolated economy, so a bad actor in one project can't drag down others.

Custodial-by-default onboarding (lift-and-shift for game players); users get keys back when they want to bridge out to Hive/Solana/etc. Two-way bridge (not just one-way).

Genesis with 3 validators (2-of-3 multisig on Hive + Solana bridge accounts), speaker becomes a regular validator after 6 months. DevNet → Testnet → Mainnet planned; this week + next is heavy testing.

blocktrades offered passkey-derived wallets (already in Magi testnet) as an alternative to custodial onboarding — speaker still wants custody-on-rails for game UX.

Side discussion: speaker asked why Hive serves JSON not Borsh — claimed 30–40% space savings. blocktrades/Bartek: custom_binary ops already exist alongside custom_json (rarely used, possibly unknown), and DB-layer compression with dictionaries removes most ASCII overhead anyway. Bartek suggested the tnauk/tune library for further JSON-footprint reduction; will share link.

Magi L2

Took over Magi dev leadership; in the last 2 months: 600% PR shipping rate, 600% net code output, 1000% multi-chain code, ~900 commits, 300 merged PRs across 13 repos, ~250k insertions — ~10x on same team/budget.

Shipped largest mainnet merge to date: BTC/HBD and HBD/Hive pools live, accessible via Altera.

TSS (threshold signature scheme) custody live on mainnet — Magi is now the 4th protocol with TSS custody and native BTC swaps without wrapping.

Off-chain gossip replacing flaky RPC; concurrency hardening; TSS lib v3 upgrade.

Built ZK proofs for the EVM integration — drops validator hardware cost (no full Eth node needed, just a prover). Forked SP1 succinct repo and took ownership; reusable across Hive.

EVM integration end-to-end on testnet: ETH + USDC deposits/swaps/withdrawals work without wrapping or custodians.

Six chain integrations done in code: BTC (live), LTC, Dash, Doge, BCH, ETH/USDC (testnet).

Altera frontend: complete rewrite done. Yozen joined the team (formerly BigD frontend).

Built passkey-derived wallets, ZK email-to-email HBD transfers (parses email body for transfer instructions).

New infra: Magi Mongo indexers, Magi NFT contracts, Magi token contracts.

Incentive pendulum (Oracle auth + economic security + validator fee distribution split with LPs) in testing — final blocker before audits with BlockTrades.

Partnerships: Dash announced; BuildX privacy-token decision Tuesday (needs $100k liquidity, 50/50 HBD/their token).

20 security review sessions, ~1000 scenarios over 2 months → 50 vulns found, 15 security PRs merged, 25 reports.

Next: launch EVM + Dash + pendulum, then freeze and move to optimization + partnerships.

howo's feedback: need many more test files / benchmarks / edge-case coverage in the repos. Elbi agreed; planned for the post-feature optimization phase.

===========

That's it for this month, feel free to ask any questions in the comments.

If you like what I do please consider supporting my proposal: https://peakd.com/proposals/371